The secret to successful AI outcomes? It’s data, says Hitachi Vantara CTO Jason Hardy

In the rapidly evolving landscape of artificial intelligence, where breakthroughs in algorithms and computing power dominate headlines, a fundamental truth often gets obscured: AI is only as powerful as the data that fuels it. According to Jason Hardy, Chief Technology Officer at Hitachi Vantara, the critical differentiator between AI projects that deliver transformative value and those that fizzle out is not the sophistication of the model—it’s the quality, governance, and strategy behind the data. In a detailed exploration of his perspective, Hardy positions data not as a mere ingredient, but as the very foundation upon which all successful AI is built. Beyond the Hype: The Data-Centric Reality of AI. The popular narrative around AI often focuses on the "brain" of the operation: the neural networks, the large language models, the cutting-edge algorithms. "There is an undeniable fascination with the model itself," Hardy acknowledges. "But think of the most advanced Formula 1 engine. 

Without the right high-octane fuel, meticulously filtered and delivered with precision, that engine sputters. In our world, data is that fuel." This shift to a data-centric AI approach is crucial. A superbly architected model trained on poor, biased, or fragmented data will inevitably produce unreliable, skewed, or even harmful outputs—a phenomenon often termed "garbage in, garbage out." Hardy emphasises that enterprises leaping into AI without first assessing their data maturity are setting themselves up for costly failures. The secret, therefore, lies in mastering the entire data pipeline long before the first model is ever trained.

 The Four Pillars of AI-Ready Data Hardy breaks down the prerequisites for data that can truly empower successful AI 

outcomes into four interconnected pillars: 1. Quality and Relevance: "Accuracy in AI starts with accuracy in data," Hardy states. This goes beyond simple data cleaning. It involves ensuring data is correct, complete, contextual, and, most importantly, relevant to the specific problem the AI aims to solve. For a predictive maintenance AI in manufacturing, for instance, relevant data includes sensor readings, maintenance logs, environmental conditions, and historical failure records—all timestamped and aligned. Irrelevant data adds noise and reduces model effectiveness.

 2. Unification and Accessibility: Data silos are the arch-nemesis of enterprise AI. Many organisations have data scattered across legacy systems, cloud repositories, and operational technology (OT) environments like factory floors. "AI requires a holistic view," explains Hardy. "You cannot ask an AI to optimise a supply chain if it only sees inventory data but not logistics, weather, or supplier reliability information." Successful outcomes depend on the ability to unify and virtualise these disparate data sources, creating a coherent and accessible data fabric that allows models to learn from a complete picture.

3. Governance, Ethics, and Trust: In an era of increasing regulation (like GDPR and AI Acts) and public scrutiny, Hardy identifies robust data governance as non-negotiable. "This is about lineage, provenance, and ethical stewardship," he says. Successful AI must be built on data whose origins are known, whose usage is compliant, and which is checked for bias.  We must know what data it was trained on to ensure it is not perpetuating historical biases." Governance builds the trust required for both internal adoption and external customer acceptance.

 4. Volume and Velocity (in Context): While not every application requires "big data" in the petabyte scale, AI models need sufficient volume to learn meaningful patterns. More critically, for real-time AI—such as fraud detection or dynamic traffic routing—the velocity of data (its streaming and processing speed) is paramount. "The data strategy must match the AI's operational tempo," Hardy notes. "Batch-processed data from yesterday is useless for an AI making decisions in milliseconds.

" The Operational Imperative: DataOps and MLOps. Understanding the importance of data is one thing; operationalising it is another. Hardy stresses the need to move from static data management to dynamic data operations. This is where Data Ops and its cousin, MLOps (Machine Learning Operations), come in. Data Ops applies agile, automated practices to the data pipeline, ensuring a continuous flow of high-quality, ready-to-use data to AI models. MLOps then manages the lifecycle of the models themselves, from training and validation to deployment and monitoring. "These disciplines ensure that your AI is not a one-off science experiment," Hardy explains. "They make it a repeatable, scalable, and maintainable industrial process. The data pipeline and the AI model pipeline must be seamlessly integrated and continuously refined." Hitachi’s Unique Lens: Converging IT and OT Data Hardy brings a perspective shaped by Hitachi’s legacy as both an IT and an Operational Technology company—a maker of software and hardware, from data storage systems to railway networks and MRI machines.

 This positions Hitachi Vantara uniquely to address one of the toughest data challenges in AI: bridging the IT/OT divide. "The most profound AI outcomes often lie at the intersection of the physical and digital worlds," he says. "To predict a train's component failure, you need OT data from vibration sensors on the wheels. To schedule its repair and manage passenger logistics, you need IT data from ERP and ticketing systems. The magic happens when you unify these domains." This convergence allows AI to solve complex societal and industrial problems—improving city mobility, advancing sustainable energy, or enabling precision medicine. Strategy and Expertise Finally, Hardy cautions against viewing data as a purely technical problem. "The secret weapon is still human expertise," he asserts. Successful outcomes require clear business objectives, cross-functional collaboration between data scientists, domain experts, and business leaders, and a strategic roadmap. "Start with the business outcome, not the technology," he advises. "Define the problem you need to solve, then work backwards to identify the data required. This ensures your AI initiatives are aligned, measurable, and driven by value, not just curiosity." Conclusion: Data as the Strategic Asset.

 In Jason Hardy’s view, the path to successful AI is a return to fundamentals. In the rush to adopt generative AI and other advanced capabilities, enterprises must not overlook the bedrock upon which everything rests. "Data is the enduring strategic asset," Hardy concludes. "Algorithms will continue to evolve and become more accessible as services, but the unique, high-quality, governed data that represents your business, your customers, and your physical operations—that is your competitive market. Investing in a modern data foundation, with robust integration, governance, and operations, isn't just preparation for AI. It is the single most critical determinant of whether your AI initiatives will succeed, fail, or deliver transformative value. That is not just the secret—it’s the essential truth."